Anticipating multi-technology convergence: a machine learning approach using patent information

Abstract

Technology convergence has been the subject of many prior studies, yet most have focussed on the structural patterns of convergence between a pair of technologies rather than the dynamic aspects of multi-technology convergence. This study proposes a machine learning approach to anticipating multi-technology convergence using patent information. For this, a patent database is first constructed using the United States Patent and Trademark Office database, distinguishing the primary class from other patent classes to consider the direction of multi-technology convergence. Second, association rule mining is employed to construct technology ecology networks describing the significant structural patterns of multi-technology convergence for different time periods in the form of a primary patent class -> supplementary patent classes. Third, the technology ecology networks between the periods are compared to identify implications on the changing patterns of multi-technology convergence. Finally, link prediction analysis based on logistic regression models is utilised to provide insight into the prospects of multi-technology convergence by identifying the links to be added to or removed from the network. Based on this, we also discuss the characteristics of the proposed approach and the technological impact and uncertainty of the identified patterns of multi-technology convergence. The case of drug, bio-affecting, and body treating compositions technology is presented herein.